A new architecture combining convolutional and transformer-based networks for automatic 3D multi-organ segmentation on CT images.

PURPOSE: Deep learning-based networks have become increasingly popular in the field of medical image segmentation. The purpose of this research was to develop and optimize a new architecture for automatic segmentation of the prostate gland and normal organs in the pelvic, thoracic, and upper gastro-intestinal (GI) regions. METHODS: We developed an architecture which combines a shifted-window (Swin) transformer with a convolutional U-Net. The network includes a parallel encoder, a cross-fusion block, and a CNN-based decoder to extract local and global information and merge related features on the same scale. A skip connection is applied between the cross-fusion block and decoder to integrate low-level semantic features. Attention gates (AGs) are integrated within the CNN to suppress features in image background regions. Our network is termed "SwinAttUNet." We optimized the architecture for automatic image segmentation. Training datasets consisted of planning-CT datasets from 300 prostate cancer patients from an institutional database and 100 CT datasets from a publicly available dataset (CT-ORG). Images were linearly interpolated and resampled to a spatial resolution of (1.0 × 1.0× 1.5) mm3 . A volume patch (192 × 192 × 96) was used for training and inference, and the dataset was split into training (75%), validation (10%), and test (15%) cohorts. Data augmentation transforms were applied consisting of random flip, rotation, and intensity scaling. The loss function comprised Dice and cross-entropy equally weighted and summed. We evaluated Dice coefficients (DSC), 95th percentile Hausdorff Distances (HD95), and Average Surface Distances (ASD) between results of our network and ground truth data. RESULTS: SwinAttUNet, DSC values were 86.54 ± 1.21, 94.15 ± 1.17, and 87.15 ± 1.68% and HD95 values were 5.06 ± 1.42, 3.16 ± 0.93, and 5.54 ± 1.63 mm for the prostate, bladder, and rectum, respectively. Respective ASD values were 1.45 ± 0.57, 0.82 ± 0.12, and 1.42 ± 0.38 mm. For the lung, liver, kidneys and pelvic bones, respective DSC values were: 97.90 ± 0.80, 96.16 ± 0.76, 93.74 ± 2.25, and 89.31 ± 3.87%. Respective HD95 values were: 5.13 ± 4.11, 2.73 ± 1.19, 2.29 ± 1.47, and 5.31 ± 1.25 mm. Respective ASD values were: 1.88 ± 1.45, 1.78 ± 1.21, 0.71 ± 0.43, and 1.21 ± 1.11 mm. Our network outperformed several existing deep learning approaches using only attention-based convolutional or Transformer-based feature strategies, as detailed in the results section. CONCLUSIONS: We have demonstrated that our new architecture combining Transformer- and convolution-based features is able to better learn the local and global context for automatic segmentation of multi-organ, CT-based anatomy.

Adaptive coherent weighted averaging algorithm for enhancement of photoacoustic tomography images of brain.

One common method to improve the low signal-to-noise ratio of the photoacoustic (PA) signal generated from weak absorbers or absorbers located in deep tissue is to acquire signal multiple times from the same region and perform averaging. However, pulse-to-pulse laser fluctuations together with differences in the beam profile of the pulses create undeterministic multiple scattering processes in the tissue. This phenomenon consequently induces a spatiotemporal displacement in the PA signal samples which in turn deteriorates the effectiveness of signal averaging. Here, we present an adaptive coherent weighted averaging algorithm to adjust the locations and values of PA signal samples for more efficient signal averaging. The proposed method is evaluated in a linear array-based PA imaging setup of ex vivo sheep brain.

ChatGPT-A promising generative AI tool and its implications for cancer care.

PLAIN LANGUAGE SUMMARY: Since its launch, ChatGPT has taken the internet by storm and has the potential to be used broadly in the health care system, particularly in a setting such as medical oncology. ChatGPT is well suited to review and extract key content from records of patients with cancer, interpret next-generation sequencing reports, and offer a list of potential clinical trial options.

Adoption of Ultrasound-guided Polymer Nanocarriers in Tumor Chemoradiotherapy and Oxidation Treatment.

The research was aimed at discussing the effectiveness of ultrasound-guided polymer nanocarriers in the clinical treatment of tumors by chemoradiotherapy and oxidation treatment. Twenty female Balb/cAnN (BALB/C) mice were selected as the research objects in the experiment. These mice were set up as tumor-bearing mice, and then ultrasound-guided polymers with different doses, including polyethylene glycol-poly 2-bromoethyl methacrylate (PEG-PBEMA) (Micelle group), free small molecules called l-ascorbyl palmitate (PA) (PA group), PA-micelle micellar particles (PA-Micelle group) prepared in the research, and phosphate buffer solution (PBS) (PBS group) were adopted. Besides, the growth of mice was recorded and compared after each operation. Meanwhile, different concentrations of PA-Micelle micellar particles and free small molecules of PA were added to the breast cancer cells of mice, and the concentration changes of glutathione (GSH) were detected to test the oxidation treatment ability of this method. According to the results of the experiment, the tumor volume of mice in the PA-Micelle group prepared in the research was the smallest followed by the PA group, and the tumor volume of mice in the Micelle group was the third smallest. The mice in the PBS group had the largest tumors among mice in all four groups. In oxidation treatment, the GSH concentration of mice in the PA-Micelle group was the lowest, while the GSH concentration of mice in the PA group was almost unchanged. The results of this experiment proved that the therapeutic effect of polymer nanocarriers in tumor chemotherapy and oxidation treatment was more significant than in traditional drug treatment.

Saliva microRNA Profile in Children with and without Severe SARS-CoV-2 Infection.

Severe acute respiratory syndrome corona virus 2 (SARS-CoV-2) may impair immune modulating host microRNAs, causing severe disease. Our objectives were to determine the salivary miRNA profile in children with SARS-CoV-2 infection at presentation and compare the expression in those with and without severe outcomes. Children <18 years with SARS-CoV-2 infection evaluated at two hospitals between March 2021 and February 2022 were prospectively enrolled. Severe outcomes included respiratory failure, shock or death. Saliva microRNAs were quantified with RNA sequencing. Data on 197 infected children (severe = 45) were analyzed. Of the known human miRNAs, 1606 (60%) were measured and compared across saliva samples. There were 43 miRNAs with ≥2-fold difference between severe and non-severe cases (adjusted p-value < 0.05). The majority (31/43) were downregulated in severe cases. The largest between-group differences involved miR-4495, miR-296-5p, miR-548ao-3p and miR-1273c. These microRNAs displayed enrichment for 32 gene ontology pathways including viral processing and transforming growth factor beta and Fc-gamma receptor signaling. In conclusion, salivary miRNA levels are perturbed in children with severe COVID-19, with the majority of miRNAs being down regulated. Further studies are required to validate and determine the utility of salivary miRNAs as biomarkers of severe COVID-19.

Putting the "mi" in omics: discovering miRNA biomarkers for pediatric precision care.

In the past decade, growing interest in micro-ribonucleic acids (miRNAs) has catapulted these small, non-coding nucleic acids to the forefront of biomarker research. Advances in scientific knowledge have made it clear that miRNAs play a vital role in regulating cellular physiology throughout the human body. Perturbations in miRNA signaling have also been described in a variety of pediatric conditions-from cancer, to renal failure, to traumatic brain injury. Likewise, the number of studies across pediatric disciplines that pair patient miRNA-omics with longitudinal clinical data are growing. Analyses of these voluminous, multivariate data sets require understanding of pediatric phenotypic data, data science, and genomics. Use of machine learning techniques to aid in biomarker detection have helped decipher background noise from biologically meaningful changes in the data. Further, emerging research suggests that miRNAs may have potential as therapeutic targets for pediatric precision care. Here, we review current miRNA biomarkers of pediatric diseases and studies that have combined machine learning techniques, miRNA-omics, and patient health data to identify novel biomarkers and potential therapeutics for pediatric diseases. IMPACT: In the following review article, we summarized how recent developments in microRNA research may be coupled with machine learning techniques to advance pediatric precision care.

An uncertainty-aware deep learning architecture with outlier mitigation for prostate gland segmentation in radiotherapy treatment planning.

PURPOSE: Task automation is essential for efficient and consistent image segmentation in radiation oncology. We report on a deep learning architecture, comprising a U-Net and a variational autoencoder (VAE) for automatic contouring of the prostate gland incorporating interobserver variation for radiotherapy treatment planning. The U-Net/VAE generates an ensemble set of segmentations for each image CT slice. A novel outlier mitigation (OM) technique was implemented to enhance the model segmentation accuracy. METHODS: The primary source dataset (source_prim) consisted of 19 200 CT slices (from 300 patient planning CT image datasets) with manually contoured prostate glands. A smaller secondary source dataset (source_sec) comprised 640 CT slices (from 10 patient CT datasets), where prostate glands were segmented by 5 independent physicians on each dataset to account for interobserver variability. Data augmentation via random rotation (<5 degrees), cropping, and horizontal flipping was applied to each dataset to increase sample size by a factor of 100. A probabilistic hierarchical U-Net with VAE was implemented and pretrained using the augmented source_prim dataset for 30 epochs. Model parameters of the U-Net/VAE were fine-tuned using the augmented source_sec dataset for 100 epochs. After the first round of training, outlier contours in the training dataset were automatically detected and replaced by the most accurate contours (based on Dice similarity coefficient, DSC) generated by the model. The U-Net/OM-VAE was retrained using the revised training dataset. Metrics for comparison included DSC, Hausdorff distance (HD, mm), normalized cross-correlation (NCC) coefficient, and center-of-mass (COM) distance (mm). RESULTS: Results for U-Net/OM-VAE with outliers replaced in the training dataset versus U-Net/VAE without OM were as follows: DSC = 0.82 ± 0.01 versus 0.80 ± 0.02 (p = 0.019), HD = 9.18 ± 1.22 versus 10.18 ± 1.35 mm (p = 0.043), NCC = 0.59 ± 0.07 versus 0.62 ± 0.06, and COM = 3.36 ± 0.81 versus 4.77 ± 0.96 mm over the average of 15 contours. For the average of 15 highest accuracy contours, values were as follows: DSC = 0.90 ± 0.02 versus 0.85 ± 0.02, HD = 5.47 ± 0.02 versus 7.54 ± 1.36 mm, and COM = 1.03 ± 0.58 versus 1.46 ± 0.68 mm (p < 0.03 for all metrics). Results for the U-Net/OM-VAE with outliers removed were as follows: DSC = 0.78 ± 0.01, HD = 10.65 ± 1.95 mm, NCC = 0.46 ± 0.10, COM = 4.17 ± 0.79 mm for the average of 15 contours, and DSC = 0.88 ± 0.02, HD = 7.00 ± 1.17 mm, COM = 1.58 ± 0.63 mm for the average of 15 highest accuracy contours. All metrics for U-Net/VAE trained on the source_prim and source_sec datasets via pretraining, followed by fine-tuning, show statistically significant improvement over that trained on the source_sec dataset only. Finally, all metrics for U-Net/VAE with or without OM showed statistically significant improvement over those for the standard U-Net. CONCLUSIONS: A VAE combined with a hierarchical U-Net and an OM strategy (U-Net/OM-VAE) demonstrates promise toward capturing interobserver variability and produces accurate prostate auto-contours for radiotherapy planning. The availability of multiple contours for each CT slice enables clinicians to determine trade-offs in selecting the "best fitting" contour on each CT slice. Mitigation of outlier contours in the training dataset improves prediction accuracy, but one must be wary of reduction in variability in the training dataset.

Effect of ultrasound-guided lumbar square muscle block on stress response in patients undergoing radical gastric cancer surgery.

BACKGROUND: Radical surgery is a common treatment for patients with gastric cancer; however, it can lead to postoperative complications and intestinal barrier dysfunction. Ultrasound-guided quadratus lumborum block is often used for postoperative analgesia, but its effects on stress response and intestinal barrier function are not well understood. AIM: To investigate the effects of an ultrasound-guided quadratus lumborum block on stress response and intestinal barrier function in patients undergoing radical surgery for gastric cancer. METHODS: A total of 100 patients undergoing radical surgery for gastric cancer were randomly categorized into observation and control groups. Plasma adrenaline and cortisol levels, intestinal mucosal barrier indexes, and complication rates were compared between the two groups before, during, and 1 day after surgery. RESULTS: The observation group had significantly lower plasma adrenaline and cortisol levels during surgery and at 1 day postoperatively than that of the control group (P < 0.05). Additionally, intestinal barrier indexes (endotoxin and D-dimer) at 1 day postoperatively were significantly lower in the observation group than in the control group (P < 0.05). CONCLUSION: Ultrasound-guided quadratus lumborum block could reduce stress response, protect intestinal barrier function, and decrease the incidence of complications in patients undergoing radical surgery for gastric cancer. This technique has the potential for clinical applications.

Analysis of the Application Value of Echocardiography Combined with CK-MB, Alb, and CysC in the Prognosis Assessment of Patients with Chronic HF.

In order to evaluate the diagnostic and prognostic value of echocardiography combined with serum creatine kinase-MB (CK-MB), albumin (Alb), and cystatin C (CysC) in patients with chronic heart failure (HF), 93 patients diagnosed with chronic HF in our hospital from March 2019 to January 2020 are retrospectively analyzed and included in the HF group. Another 100 healthy subjects who come to our hospital for general physical examination are selected as the control group. Echocardiography is used to detect the cardiac parameters of each group. The experimental results show that echocardiography parameters combined with CK-MB, Alb, and CysC have high application value in diagnosis and evaluation of patients with chronic HF, which can provide theoretical basis for improving the prognosis of patients with chronic HF through real-time monitoring of the above indicators.

Hemodynamic profiles by non-invasive monitoring of cardiac index and vascular tone in acute heart failure patients in the emergency department: External validation and clinical outcomes.

BACKGROUND: Non-invasive finger-cuff monitors measuring cardiac index and vascular tone (SVRI) classify emergency department (ED) patients with acute heart failure (AHF) into three otherwise-indistinguishable subgroups. Our goals were to validate these "hemodynamic profiles" in an external cohort and assess their association with clinical outcomes. METHODS: AHF patients (n = 257) from five EDs were prospectively enrolled in the validation cohort (VC). Cardiac index and SVRI were measured with a ClearSight finger-cuff monitor (formerly NexFin, Edwards Lifesciences) as in a previous study (derivation cohort, DC, n = 127). A control cohort (CC, n = 127) of ED patients with sepsis was drawn from the same study as the DC. K-means cluster analysis previously derived two-dimensional (cardiac index and SVRI) hemodynamic profiles in the DC and CC (k = 3 profiles each). The VC was subgrouped de novo into three analogous profiles by unsupervised K-means consensus clustering. PERMANOVA tested whether VC profiles 1-3 differed from profiles 1-3 in the DC and CC, by multivariate group composition of cardiac index and vascular tone. Profiles in the VC were compared by a primary outcome of 90-day mortality and a 30-day ranked composite secondary outcome (death, mechanical cardiac support, intubation, new/emergent dialysis, coronary intervention/surgery) as time-to-event (survival analysis) and binary events (odds ratio, OR). Descriptive statistics were used to compare profiles by two validated risk scores for the primary outcome, and one validated score for the secondary outcome. RESULTS: The VC had median age 60 years (interquartile range {49-67}), and was 45% (n = 116) female. Multivariate profile composition by cardiac index and vascular tone differed significantly between VC profiles 1-3 and CC profiles 1-3 (p = 0.001, R2 = 0.159). A difference was not detected between profiles in the VC vs. the DC (p = 0.59, R2 = 0.016). VC profile 3 had worse 90-day survival than profiles 1 or 2 (HR = 4.8, 95%CI 1.4-17.1). The ranked secondary outcome was more likely in profile 1 (OR = 10.0, 1.2-81.2) and profile 3 (12.8, 1.7-97.9) compared to profile 2. Diabetes prevalence and blood urea nitrogen were lower in the high-risk profile 3 (p<0.05). No significant differences between profiles were observed for other clinical variables or the 3 clinical risk scores. CONCLUSIONS: Hemodynamic profiles in ED patients with AHF, by non-invasive finger-cuff monitoring of cardiac index and vascular tone, were replicated de novo in an external cohort. Profiles showed significantly different risks of clinically-important adverse patient outcomes.

Diagnostic role of contrast-enhanced ultrasonography versus conventional B-mode ultrasonography in cirrhotic patients with early hepatocellular carcinoma: a retrospective study.

BACKGROUND: Early diagnosis of hepatocellular carcinoma (HCC) is very important for the prognosis of patients. However, there are very few studies that compared the diagnostic accuracy of contrast-enhanced ultrasonography (CEUS) and B-mode ultrasonography for early HCC in cirrhotic patients. METHODS: This retrospective study included cirrhotic patients, who were suspected of early HCC between January 2020 and June 2021. The diagnosis of patients was based on the pathology results of surgery or biopsy. Demographic and clinical characteristics of included patients were recorded. The diagnoses of suspected lesions using both types of ultrasonography were recorded, and the sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and diagnostic accuracy of early HCC in cirrhotic patients were calculated. RESULTS: Eventually, 137 patients with solitary lesions in the liver were included in this study, including 89 patients diagnosed with HCC and 48 patients diagnosed with non-HCC. The median diameter of suspected lesions was 26 mm, and the median level of alpha fetoprotein (AFP) was 37.2 ng/mL. When comparing the demographic and clinical characteristics of cirrhotic patients with HCC and non-HCC, it was found that patients with HCC had significantly higher levels of AFP than those with non-HCC (P=0.03). The sensitivity, specificity, PPV, NPV, and accuracy of CEUS in early HCC were 73%, 93.8%, 95.6%, 65.2% and 80.3%, respectively. In CEUS, all of these parameters were much higher than those in B-mode ultrasonography, i.e., 64%, 75%, 82.6%, 52.9%, and 67.9%. It was also found that the diagnostic accuracy of CEUS was much higher than that of B-mode ultrasonography especially regarding lesions <20 mm. To further improve the sensitivity of CEUS in early HCC, AFP was combined with CEUS for the diagnosis of early HCC. As a result, the sensitivity, specificity, PPV, NPV, and accuracy of CEUS combined with AFP level were 83.1%, 87.5%, 92.5%, 73.7%, and 84.7%, respectively. CONCLUSIONS: Our study confirmed that CEUS' diagnostic accuracy for early HCC in cirrhotic patients was significantly higher than that of B-mode ultrasonography. However, the sensitivity of CEUS needs to be improved further, and the combination of CEUS and AFP level may be a potential solution.

Application of velocity vector imaging (VVI) in the evaluation of left ventricular myocardial contractile strain and dyssynchrony before and after stent implantation in patients with coronary heart disease.

BACKGROUND: Percutaneous coronary intervention (PCI) has become increasingly mature and has gradually become the main treatment for coronary heart disease (CHD). However, evaluation of myocardial reperfusion after PCI remains a major clinical challenge. This study aimed to explore the VVI technique in evaluating the effect, prognosis, and follow-up of CHD patients after percutaneous coronary intervention. We performed a quantitative analysis of left ventricular myocardial contractile strain and dyssynchrony before and after stent implantation in patients by VVI. METHODS: Thirty-five patients diagnosed with CHD who underwent percutaneous coronary stenting (PCI) in the Department of Cardiovascular Medicine, Affiliated Hospital of Jiangnan University from March 2019 to October 2020 were selected as the case group. Continuous dynamic two-dimensional images of the patient's left ventricle were analyzed using VVI at 1 day before PCI (group A), 7 days after PCI (group B), and 30 days after PCI (group C). The patients' left ventricular end diastolic diameter (LVEDD), left ventricular end systolic diameter (LVESD), left ventricle ejection fraction (LVEF), peak longitudinal strain, and peak radial strain of myocardial contraction were measured. The VVI images of 35 healthy subjects who underwent physical examination in the outpatient department of our hospital from March 2019 to October 2020 were selected as controls. RESULTS: There were no significant differences in the LVEF, LVEDD, and LVESD between the case and control groups (P>0.05). The peak systolic longitudinal and radial strain values at 1 month after treatment were higher than those before treatment. The differences among myocardial segments were statistically significant, except for the apical septum, base anterior, apical anterior, and base inferior segments (P<0.05). The peak systolic longitudinal and radial strain values at 1 week after treatment were not significantly different from those at 1 month after treatment, except for the base anterior septum, mid anterior, posterior, and inferior myocardial segments (P>0.05). CONCLUSIONS: VVI technology can comprehensively and objectively evaluate the overall and local myocardial function of the left ventricle, thereby providing a novel method for the clinical treatment of CHD as well as the evaluation of curative effect and prognosis.

Tackling Ordinal Regression Problem for Heterogeneous Data: Sparse and Deep Multi-Task Learning Approaches.

Many real-world datasets are labeled with natural orders, i.e., ordinal labels. Ordinal regression is a method to predict ordinal labels that finds a wide range of applications in data-rich domains, such as natural, health and social sciences. Most existing ordinal regression approaches work well for independent and identically distributed (IID) instances via formulating a single ordinal regression task. However, for heterogeneous non-IID instances with well-defined local geometric structures, e.g., subpopulation groups, multi-task learning (MTL) provides a promising framework to encode task (subgroup) relatedness, bridge data from all tasks, and simultaneously learn multiple related tasks in efforts to improve generalization performance. Even though MTL methods have been extensively studied, there is barely existing work investigating MTL for heterogeneous data with ordinal labels. We tackle this important problem via sparse and deep multi-task approaches. Specifically, we develop a regularized multi-task ordinal regression (MTOR) model for smaller datasets and a deep neural networks based MTOR model for large-scale datasets. We evaluate the performance using three real-world healthcare datasets with applications to multi-stage disease progression diagnosis. Our experiments indicate that the proposed MTOR models markedly improve the prediction performance comparing with single-task ordinal regression models.

Effect of upper gastrointestinal bleeding on prognosis of middle-aged patients with acute ischemic stroke: a retrospective study.

BACKGROUND: Upper gastrointestinal bleeding (UGIB) is a common complication of acute ischemic stroke (AIS), but the effect of UGIB on the prognosis of middle-aged AIS patients is not clear. METHODS: Patients with AIS admitted to our hospital from January 2011 to December 2015 were eligible to be included in this study. All included patients were divided into UGIB and non-UGIB groups. Some clinical characteristics were retrospectively collected. Primary outcomes were all-cause mortality within 1, 3, and 5 years, as well as the incidence of stroke recurrence. Cox proportional hazards regression analyses were used to determine the effect of UGIB on 5-year mortality and the incidence of stroke recurrence. Logistic regression was also used to identify the predictors of UGIB in AIS patients. RESULTS: A total of 405 AIS patients were included in this study and then divided into UGIB and non-UGIB groups. The mean age of the UGIB group and non-UGIB group was 61.5±9.6 and 53.1±14.0 years, respectively (P<0.001). The baseline score of the National Institute of Health Stroke Scale (NIHSS) was significantly higher in the UGIB group than in the non-UGIB group (P<0.001). AIS patients in the UGIB group had a higher 1-, 3-, and 5-year mortality and a higher incidence of stroke recurrence (all P<0.001). Kaplan-Meier curves showed that AIS patients with UGIB had a higher 5-year mortality and a higher incidence of stroke recurrence (both P<0.001). Cox proportional hazards regression models indicated that the occurrence of UGIB, older age, a high NIHSS score, and stroke recurrence were related to a higher 5-year mortality. Similarly, the occurrence of UGIB, older age, a high NIHSS score, and hypertension increased the incidence of stroke recurrence. According to the multivariate logistic regression analysis, older age, a high NIHSS score, and previous anticoagulant use were identified as predictors of UGIB. CONCLUSIONS: UGIB has important effects on the prognosis of AIS patients. The incidence of UGIB increases with older age, a high NIHSS score, and previous anticoagulant use, which provides important evidence for the treatment and nursing of AIS patients.

The Population Health OutcomEs aNd Information EXchange (PHOENIX) Program - A Transformative Approach to Reduce the Burden of Chronic Disease.

This concept article introduces a transformative vision to reduce the population burden of chronic disease by focusing on data integration, analytics, implementation and community engagement. Known as PHOENIX (The Population Health OutcomEs aNd Information EXchange), the approach leverages a state level health information exchange and multiple other resources to facilitate the integration of clinical and social determinants of health data with a goal of achieving true population health monitoring and management. After reviewing historical context, we describe how multilevel and multimodal data can be used to facilitate core public health services, before discussing the controversies and challenges that lie ahead.

Predicting Clinical Outcomes with Patient Stratification via Deep Mixture Neural Networks.

The increasing availability of electronic health record data offers unprecedented opportunities for predictive modeling in healthcare informatics including outcomes such as mortality and disease diagnosis as well as risk factor identification. Recently, deep neural networks (DNNs) have been successfully applied in healthcare informatics and achieved state-of-art predictive performance. However, existing DNN models either rely on the pre-defined patient subgroups or take the "one-size-fits-all" approach and are built without considering patient stratification. Consequently, those models are not able to discover patient subgroups and the risk factors are thereafter identified for the entire patient population, failing to account for potential group differences. To address this challenge, we propose the use of deep mixture neural networks (DMNN), a unified DNN model for simultaneous patient stratification and predictive modeling. Experimental results on a clinic dataset show that our proposed DMNN can achieve good performance on predicting diagnosis of acute heart failure. With DMNN's ability to incorporate patient stratification, we are able to systematically identify group-specific risk factors for different patient subgroups which could potentially shed light on revealing factors that contribute to outcome differences.

Deep learning protocol for improved photoacoustic brain imaging.

One of the key limitations for the clinical translation of photoacoustic imaging is penetration depth that is linked to the tissue maximum permissible exposures (MPE) recommended by the American National Standards Institute (ANSI). Here, we propose a method based on deep learning to virtually increase the MPE in order to enhance the signal-to-noise ratio of deep structures in the brain tissue. The proposed method is evaluated in an in vivo sheep brain imaging experiment. We believe this method can facilitate clinical translation of photoacoustic technique in brain imaging, especially in transfontanelle brain imaging in neonates.

Leveraging auxiliary measures: a deep multi-task neural network for predictive modeling in clinical research.

BACKGROUND: Accurate predictive modeling in clinical research enables effective early intervention that patients are most likely to benefit from. However, due to the complex biological nature of disease progression, capturing the highly non-linear information from low-level input features is quite challenging. This requires predictive models with high-capacity. In practice, clinical datasets are often of limited size, bringing danger of overfitting for high-capacity models. To address these two challenges, we propose a deep multi-task neural network for predictive modeling. METHODS: The proposed network leverages clinical measures as auxiliary targets that are related to the primary target. The predictions for the primary and auxiliary targets are made simultaneously by the neural network. Network structure is specifically designed to capture the clinical relevance by learning a shared feature representation between the primary and auxiliary targets. We apply the proposed model in a hypertension dataset and a breast cancer dataset, where the primary tasks are to predict the left ventricular mass indexed to body surface area and the time of recurrence of breast cancer. Moreover, we analyze the weights of the proposed neural network to rank input features for model interpretability. RESULTS: The experimental results indicate that the proposed model outperforms other different models, achieving the best predictive accuracy (mean squared error 199.76 for hypertension data, 860.62 for Wisconsin prognostic breast cancer data) with the ability to rank features according to their contributions to the targets. The ranking is supported by previous related research. CONCLUSION: We propose a novel effective method for clinical predictive modeling by combing the deep neural network and multi-task learning. By leveraging auxiliary measures clinically related to the primary target, our method improves the predictive accuracy. Based on featue ranking, our model is interpreted and shows consistency with previous studies on cardiovascular diseases and cancers.

Hypoxia-Inducible Factor 1 alpha (HIF-1α)/Vascular Endothelial Growth Factor (VEGF) Pathway Participates in Angiogenesis of Myocardial Infarction in Muscone-Treated Mice: Preliminary Study.

BACKGROUND Angiogenesis plays a crucial role in myocardial infarction (MI) treatment by ameliorating myocardial remodeling, thus improving cardiac function and preventing heart failure. Muscone has been reported to have beneficial effects on cardiac remodeling in MI mice. However, the effects of muscone on angiogenesis in MI mice and its underlying mechanisms remain unknown. MATERIAL AND METHODS Mice were randomly divided into sham, MI, and MI+muscone groups. The MI mouse model was established by ligating the left anterior descending coronary artery. Mice in the sham group received the same procedure except for ligation. Mice were administered muscone or an equivalent volume of saline for 4 consecutive weeks. Cardiac function was evaluated by echocardiograph after MI for 2 and 4 weeks. Four weeks later, all mice were sacrificed and Masson's trichrome staining was used to assess myocardial fibrosis. Isolectin B4 staining was applied to evaluate the angiogenesis in mouse hearts. Immunohistochemistry, Western blot analysis, and quantitative real-time polymerase chain reaction (qPCR) were performed to analyze expression levels of HIF-1a and its downstream genes. RESULTS Compared with the MI group, muscone treatment significantly improved cardiac function and reduced myocardial fibrosis. Moreover, muscone enhanced angiogenesis in the peri-infarct region and p-VEGFR2 expression in the vascular endothelial cells. Western blot analysis and qPCR showed that muscone upregulated expression levels of HIF-1a and VEGFA. CONCLUSIONS Muscone improved cardiac function in MI mice through augmented angiogenesis. The potential mechanism of muscone treatment in regulating angiogenesis of MI mice was upregulating expression levels of HIF-1α and VEGFA.